search

NetEase-GameAI / ProPose

[CVPR 2023] Learning Analytical Posterior Probability for Human Mesh Recovery
BSD 3-Clause "New" or "Revised" License
137 stars 9 forks source link

About the SMPL annotations of h36m dataset #6

Closed XiaobenLi00 closed 1 year ago

XiaobenLi00 commented 1 year ago

I know h36m is not a open source dataset, but I can only get 3D labels, could you provide some information about how to get SMPL annotations of it?

raypine commented 1 year ago

If the MoSh data is not available, you can follow the data process of HybrIK (this section) and he gets the data from Moon.

XiaobenLi00 commented 1 year ago

Thanks a lot for your reply. But I still wonder how you get the GT SMPL parameters of h36m, did you apply Mosh to the marker data or you get them somewhere?

XiaobenLi00 commented 1 year ago

Thanks a lot for your reply. But I still wonder how you get the GT SMPL parameters of h36m, did you apply Mosh to the marker data or you get them somewhere?

So did you follow the data process of HybrIK and get the data from Moon?

XiaobenLi00 commented 1 year ago

If the MoSh data is not available, you can follow the data process of HybrIK (this section) and he gets the data from Moon.

I check the annoation data from Moon, but it did not contain SMPL parameters right?

raypine commented 1 year ago

Actually It has the parameters in annotations (Human36M_subjectX_smpl_param.json).

raypine commented 1 year ago

Or you can just download the parsed annotations from Hybrik with the link.

XiaobenLi00 commented 1 year ago

Or you can just download the parsed annotations from Hybrik with the link.

Yes, I can find annotations in this link.

XiaobenLi00 commented 1 year ago

Actually It has the parameters in annotations (Human36M_subjectX_smpl_param.json).

But sorry for this, I cannot find *smpl_param.json in the 'annotations'

image
raypine commented 1 year ago

Oh..., I guess Moon has modified this file. Nevertheless, the data from HybrIK is still complete, right ?

XiaobenLi00 commented 1 year ago

Oh..., I guess Moon has modified this file. Nevertheless, the data from HybrIK is still complete, right ?

These are the data from HybrIK

image

And I find a data list in other repository of MoonI2L-MeshNet, the data link is smpl and the data is as follows

image

So are these data indeed the 'Human36M_subjectX_smpl_param.json' as you said before? Thanks again for your patience :)

raypine commented 1 year ago

Yes, that's right. I2L-meshnet uses the SMPL and posenet only uses 3D keypoints.

XiaobenLi00 commented 1 year ago

Yes, that's right. I2L-meshnet uses the SMPL and posenet only uses 3D keypoints.

Thanks a lot!